说明：是用零码自训练语音识别语言模型，声音分类模型。提高业务领域专有名词识别准确率，区分不同声音类别，广泛应用于行业数据采集录入、语音指令、呼叫中心、声音类型检测等应用场景。说白了，就是基于语音识别上的牛杯功能。
[En]
Explain: is to use zero code self-training speech recognition language model, sound classification model. Improve the accuracy of business domain proper noun recognition and distinguish different sound categories, which is widely used in industry data acquisition and input, voice instruction, call center, sound type detection and other application scenarios. To put it bluntly, it is based on the Bull Cup function on speech recognition.
有四个一般的培训程序。语音识别可以利用商业场景文本的语料库自动训练语音识别模型，而无需用户键入代码，从而提高了商业领域识别的准确性。一般来说，更常见的场景有语音对话、语音指令(小)、语音输入(微信语音)、客服电话(超级烦人的机器人只是不会把你转到人工0 0)等。
[En]
There are four general training procedures. Speech recognition can use the corpus of the business scene text to automatically train the speech recognition model without the user typing the code, so as to improve the accuracy of business domain recognition. Generally speaking, the more common scenes are in voice conversation, voice instructions (small), voice input (Wechat voice), customer service calls (super annoying robots just don’t transfer you to manual 0 0), and so on.
- 1.创建模型，选择训练的语音识别接口
- 2.上传测试音频和标注文本评估基础模型
- 3.上传业务词汇或长文本自动训练模型
- 4.上线模型，语音识别接口配置参数使用
说白了，就是不用写代码、准备声音素材，也不用上传。然后将这些声音对应的文字或文字发送给他进行训练，当然也要及时调整控制训练结果。然后完成训练，整个模型就可以投入使用了。也就是说，当他听到或识别出类似的发音时，他可以反映出文本是什么。挺好玩的。
[En]
To put it bluntly, you don’t have to write code, prepare the sound material and upload it. Then send the text or words corresponding to these voices to give him training, of course, you have to adjust the control training results in time. Then the training is completed and the whole model can be put into use. That is, when he hears or recognizes a similar pronunciation, he can reflect what the text is. It was fun.
我亲测过让机器识别 哈比 憨批 憨憨 哈子 艾斯比。。。结果机器都能识别出来是啥。简直优秀。
除了语音识别之外的另一个类别是语音分类。
[En]
Another category in addition to speech recognition is voice classification.
核心内容是定制和识别当前音频是什么类型的声音。用于监控生产或泛安全场景中的异常声音。它被广泛应用于安全监控和科学研究。
[En]
The core content is to customize and identify what type of sound the current audio is. It is used to monitor abnormal sounds in production or pan-security scenarios. It is widely used in security monitoring and scientific research.
该流程类似于上面的语音识别，总共有四个步骤：
[En]
The process is similar to the speech recognition above, with a total of four:
- 数据处理 提供闭环的数据管理功能，从数据上传、标注到训练
- 模型训练 提供丰富的训练方式，零代码轻松获得高精度模型
- 模型校验 提供详细的模型评估报告，支持在线校验，助力针对性优化模型
- 模型部署 将模型转换为适合业务场景的推理形式，从云到端全覆盖
一般以上的语音识别功能就这些，然后稍微讲讲百度的EasyDL语音识别。和其他云厂商的有一些不同和优势，主要体现在：
最快 10min训练优化
一站式自动化训练
上传文件极简交互
可视化训练报告
系统自动评价推荐基本模型
[En]
System automatic evaluation recommendation basic model
训练前后均提供评估报告
5%-25%识别率提升
预置百度大规模预训模式
[En]
Preset Baidu large-scale pre-training model
支持多种长短文本训练方式
[En]
Support multiple training methods for long and short texts
支持多次上传迭代训练
多种云端调用方式
模型上线后专属使用
支持在线API，SDK多种方式
Original: https://blog.csdn.net/m0_66194642/article/details/124248884
Author: 打工人何苦为难打工人
Title: 云云云云云云云云EasyDL 语音（3）
原创文章受到原创版权保护。转载请注明出处：https://www.johngo689.com/497922/
转载文章受原作者版权保护。转载请注明原作者出处！